An Approach to Analyzing Emotions in Diary Text Entries

This article introduces a hybrid approach for the classification of emotions in Russian dream diary texts. The research focuses on processing semi-structured data characterized by a high subjectivity in describing real-world entities. In this context, traditional approaches often show limited effectiveness. The article analyzes contemporary techniques for tonality and emotion analysis, highlighting their limitations when applied to the subjective nature of dream content. The proposed approach integrates lexical methods, machine learning algorithms, and transformer-based models to enhance performance. The result of the article is also an annotated dataset, which includes labels for six basic emotions, according to the Paul Ekman classification scheme. Experimental results show that the hybrid model achieves a precision of 22% in classifying two emotions simultaneously. It also achieves a 56% precision in classifying a single emotion. Overall, it outperforms comparable methods. We conducted an ablation study to validate the contribution of each component within the hybrid system, confirming their significance for overall performance.